2 * PCI Error Recovery Driver for RPA-compliant PPC64 platform.
3 * Copyright IBM Corp. 2004 2005
4 * Copyright Linas Vepstas <linas@linas.org> 2004, 2005
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License as published by
10 * the Free Software Foundation; either version 2 of the License, or (at
11 * your option) any later version.
13 * This program is distributed in the hope that it will be useful, but
14 * WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
16 * NON INFRINGEMENT. See the GNU General Public License for more
19 * You should have received a copy of the GNU General Public License
20 * along with this program; if not, write to the Free Software
21 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
23 * Send comments and feedback to Linas Vepstas <linas@austin.ibm.com>
25 #include <linux/delay.h>
26 #include <linux/interrupt.h>
27 #include <linux/irq.h>
28 #include <linux/module.h>
29 #include <linux/pci.h>
31 #include <asm/eeh_event.h>
32 #include <asm/ppc-pci.h>
33 #include <asm/pci-bridge.h>
38 * eeh_pcid_name - Retrieve name of PCI device driver
41 * This routine is used to retrieve the name of PCI device driver
44 static inline const char *eeh_pcid_name(struct pci_dev *pdev)
46 if (pdev && pdev->dev.driver)
47 return pdev->dev.driver->name;
52 * eeh_pcid_get - Get the PCI device driver
55 * The function is used to retrieve the PCI device driver for
56 * the indicated PCI device. Besides, we will increase the reference
57 * of the PCI device driver to prevent that being unloaded on
58 * the fly. Otherwise, kernel crash would be seen.
60 static inline struct pci_driver *eeh_pcid_get(struct pci_dev *pdev)
62 if (!pdev || !pdev->driver)
65 if (!try_module_get(pdev->driver->driver.owner))
72 * eeh_pcid_put - Dereference on the PCI device driver
75 * The function is called to do dereference on the PCI device
76 * driver of the indicated PCI device.
78 static inline void eeh_pcid_put(struct pci_dev *pdev)
80 if (!pdev || !pdev->driver)
83 module_put(pdev->driver->driver.owner);
87 * eeh_disable_irq - Disable interrupt for the recovering device
90 * This routine must be called when reporting temporary or permanent
91 * error to the particular PCI device to disable interrupt of that
92 * device. If the device has enabled MSI or MSI-X interrupt, we needn't
93 * do real work because EEH should freeze DMA transfers for those PCI
94 * devices encountering EEH errors, which includes MSI or MSI-X.
96 static void eeh_disable_irq(struct pci_dev *dev)
98 struct eeh_dev *edev = pci_dev_to_eeh_dev(dev);
100 /* Don't disable MSI and MSI-X interrupts. They are
101 * effectively disabled by the DMA Stopped state
102 * when an EEH error occurs.
104 if (dev->msi_enabled || dev->msix_enabled)
107 if (!irq_has_action(dev->irq))
110 edev->mode |= EEH_DEV_IRQ_DISABLED;
111 disable_irq_nosync(dev->irq);
115 * eeh_enable_irq - Enable interrupt for the recovering device
118 * This routine must be called to enable interrupt while failed
119 * device could be resumed.
121 static void eeh_enable_irq(struct pci_dev *dev)
123 struct eeh_dev *edev = pci_dev_to_eeh_dev(dev);
125 if ((edev->mode) & EEH_DEV_IRQ_DISABLED) {
126 edev->mode &= ~EEH_DEV_IRQ_DISABLED;
130 * This is just ass backwards. This maze has
131 * unbalanced irq_enable/disable calls. So instead of
132 * finding the root cause it works around the warning
133 * in the irq_enable code by conditionally calling
136 * That's just wrong.The warning in the core code is
137 * there to tell people to fix their assymetries in
138 * their own code, not by abusing the core information
141 * I so wish that the assymetry would be the other way
142 * round and a few more irq_disable calls render that
143 * shit unusable forever.
147 if (irqd_irq_disabled(irq_get_irq_data(dev->irq)))
148 enable_irq(dev->irq);
152 static bool eeh_dev_removed(struct eeh_dev *edev)
154 /* EEH device removed ? */
155 if (!edev || (edev->mode & EEH_DEV_REMOVED))
161 static void *eeh_dev_save_state(void *data, void *userdata)
163 struct eeh_dev *edev = data;
164 struct pci_dev *pdev;
169 pdev = eeh_dev_to_pci_dev(edev);
173 pci_save_state(pdev);
178 * eeh_report_error - Report pci error to each device driver
180 * @userdata: return value
182 * Report an EEH error to each device driver, collect up and
183 * merge the device driver responses. Cumulative response
184 * passed back in "userdata".
186 static void *eeh_report_error(void *data, void *userdata)
188 struct eeh_dev *edev = (struct eeh_dev *)data;
189 struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
190 enum pci_ers_result rc, *res = userdata;
191 struct pci_driver *driver;
193 if (!dev || eeh_dev_removed(edev))
195 dev->error_state = pci_channel_io_frozen;
197 driver = eeh_pcid_get(dev);
198 if (!driver) return NULL;
200 eeh_disable_irq(dev);
202 if (!driver->err_handler ||
203 !driver->err_handler->error_detected) {
208 rc = driver->err_handler->error_detected(dev, pci_channel_io_frozen);
210 /* A driver that needs a reset trumps all others */
211 if (rc == PCI_ERS_RESULT_NEED_RESET) *res = rc;
212 if (*res == PCI_ERS_RESULT_NONE) *res = rc;
219 * eeh_report_mmio_enabled - Tell drivers that MMIO has been enabled
221 * @userdata: return value
223 * Tells each device driver that IO ports, MMIO and config space I/O
224 * are now enabled. Collects up and merges the device driver responses.
225 * Cumulative response passed back in "userdata".
227 static void *eeh_report_mmio_enabled(void *data, void *userdata)
229 struct eeh_dev *edev = (struct eeh_dev *)data;
230 struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
231 enum pci_ers_result rc, *res = userdata;
232 struct pci_driver *driver;
234 if (!dev || eeh_dev_removed(edev))
237 driver = eeh_pcid_get(dev);
238 if (!driver) return NULL;
240 if (!driver->err_handler ||
241 !driver->err_handler->mmio_enabled ||
242 (edev->mode & EEH_DEV_NO_HANDLER)) {
247 rc = driver->err_handler->mmio_enabled(dev);
249 /* A driver that needs a reset trumps all others */
250 if (rc == PCI_ERS_RESULT_NEED_RESET) *res = rc;
251 if (*res == PCI_ERS_RESULT_NONE) *res = rc;
258 * eeh_report_reset - Tell device that slot has been reset
260 * @userdata: return value
262 * This routine must be called while EEH tries to reset particular
263 * PCI device so that the associated PCI device driver could take
264 * some actions, usually to save data the driver needs so that the
265 * driver can work again while the device is recovered.
267 static void *eeh_report_reset(void *data, void *userdata)
269 struct eeh_dev *edev = (struct eeh_dev *)data;
270 struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
271 enum pci_ers_result rc, *res = userdata;
272 struct pci_driver *driver;
274 if (!dev || eeh_dev_removed(edev))
276 dev->error_state = pci_channel_io_normal;
278 driver = eeh_pcid_get(dev);
279 if (!driver) return NULL;
283 if (!driver->err_handler ||
284 !driver->err_handler->slot_reset ||
285 (edev->mode & EEH_DEV_NO_HANDLER)) {
290 rc = driver->err_handler->slot_reset(dev);
291 if ((*res == PCI_ERS_RESULT_NONE) ||
292 (*res == PCI_ERS_RESULT_RECOVERED)) *res = rc;
293 if (*res == PCI_ERS_RESULT_DISCONNECT &&
294 rc == PCI_ERS_RESULT_NEED_RESET) *res = rc;
300 static void *eeh_dev_restore_state(void *data, void *userdata)
302 struct eeh_dev *edev = data;
303 struct pci_dev *pdev;
308 pdev = eeh_dev_to_pci_dev(edev);
312 pci_restore_state(pdev);
317 * eeh_report_resume - Tell device to resume normal operations
319 * @userdata: return value
321 * This routine must be called to notify the device driver that it
322 * could resume so that the device driver can do some initialization
323 * to make the recovered device work again.
325 static void *eeh_report_resume(void *data, void *userdata)
327 struct eeh_dev *edev = (struct eeh_dev *)data;
328 struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
329 struct pci_driver *driver;
331 if (!dev || eeh_dev_removed(edev))
333 dev->error_state = pci_channel_io_normal;
335 driver = eeh_pcid_get(dev);
336 if (!driver) return NULL;
340 if (!driver->err_handler ||
341 !driver->err_handler->resume ||
342 (edev->mode & EEH_DEV_NO_HANDLER)) {
343 edev->mode &= ~EEH_DEV_NO_HANDLER;
348 driver->err_handler->resume(dev);
355 * eeh_report_failure - Tell device driver that device is dead.
357 * @userdata: return value
359 * This informs the device driver that the device is permanently
360 * dead, and that no further recovery attempts will be made on it.
362 static void *eeh_report_failure(void *data, void *userdata)
364 struct eeh_dev *edev = (struct eeh_dev *)data;
365 struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
366 struct pci_driver *driver;
368 if (!dev || eeh_dev_removed(edev))
370 dev->error_state = pci_channel_io_perm_failure;
372 driver = eeh_pcid_get(dev);
373 if (!driver) return NULL;
375 eeh_disable_irq(dev);
377 if (!driver->err_handler ||
378 !driver->err_handler->error_detected) {
383 driver->err_handler->error_detected(dev, pci_channel_io_perm_failure);
389 static void *eeh_rmv_device(void *data, void *userdata)
391 struct pci_driver *driver;
392 struct eeh_dev *edev = (struct eeh_dev *)data;
393 struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
394 int *removed = (int *)userdata;
397 * Actually, we should remove the PCI bridges as well.
398 * However, that's lots of complexity to do that,
399 * particularly some of devices under the bridge might
400 * support EEH. So we just care about PCI devices for
403 if (!dev || (dev->hdr_type & PCI_HEADER_TYPE_BRIDGE))
407 * We rely on count-based pcibios_release_device() to
408 * detach permanently offlined PEs. Unfortunately, that's
409 * not reliable enough. We might have the permanently
410 * offlined PEs attached, but we needn't take care of
411 * them and their child devices.
413 if (eeh_dev_removed(edev))
416 driver = eeh_pcid_get(dev);
419 if (driver->err_handler &&
420 driver->err_handler->error_detected &&
421 driver->err_handler->slot_reset &&
422 driver->err_handler->resume)
426 /* Remove it from PCI subsystem */
427 pr_debug("EEH: Removing %s without EEH sensitive driver\n",
429 edev->bus = dev->bus;
430 edev->mode |= EEH_DEV_DISCONNECTED;
433 pci_lock_rescan_remove();
434 pci_stop_and_remove_bus_device(dev);
435 pci_unlock_rescan_remove();
440 static void *eeh_pe_detach_dev(void *data, void *userdata)
442 struct eeh_pe *pe = (struct eeh_pe *)data;
443 struct eeh_dev *edev, *tmp;
445 eeh_pe_for_each_dev(pe, edev, tmp) {
446 if (!(edev->mode & EEH_DEV_DISCONNECTED))
449 edev->mode &= ~(EEH_DEV_DISCONNECTED | EEH_DEV_IRQ_DISABLED);
450 eeh_rmv_from_parent_pe(edev);
457 * Explicitly clear PE's frozen state for PowerNV where
458 * we have frozen PE until BAR restore is completed. It's
459 * harmless to clear it for pSeries. To be consistent with
460 * PE reset (for 3 times), we try to clear the frozen state
461 * for 3 times as well.
463 static void *__eeh_clear_pe_frozen_state(void *data, void *flag)
465 struct eeh_pe *pe = (struct eeh_pe *)data;
466 bool *clear_sw_state = flag;
469 for (i = 0; rc && i < 3; i++)
470 rc = eeh_unfreeze_pe(pe, clear_sw_state);
472 /* Stop immediately on any errors */
474 pr_warn("%s: Failure %d unfreezing PHB#%x-PE#%x\n",
475 __func__, rc, pe->phb->global_number, pe->addr);
482 static int eeh_clear_pe_frozen_state(struct eeh_pe *pe,
487 rc = eeh_pe_traverse(pe, __eeh_clear_pe_frozen_state, &clear_sw_state);
489 eeh_pe_state_clear(pe, EEH_PE_ISOLATED);
491 return rc ? -EIO : 0;
494 int eeh_pe_reset_and_recover(struct eeh_pe *pe)
498 /* Bail if the PE is being recovered */
499 if (pe->state & EEH_PE_RECOVERING)
502 /* Put the PE into recovery mode */
503 eeh_pe_state_mark(pe, EEH_PE_RECOVERING);
506 eeh_pe_dev_traverse(pe, eeh_dev_save_state, NULL);
509 eeh_pe_dev_traverse(pe, eeh_report_error, &result);
512 ret = eeh_reset_pe(pe);
514 eeh_pe_state_clear(pe, EEH_PE_RECOVERING);
518 /* Unfreeze the PE */
519 ret = eeh_clear_pe_frozen_state(pe, true);
521 eeh_pe_state_clear(pe, EEH_PE_RECOVERING);
525 /* Notify completion of reset */
526 eeh_pe_dev_traverse(pe, eeh_report_reset, &result);
528 /* Restore device state */
529 eeh_pe_dev_traverse(pe, eeh_dev_restore_state, NULL);
532 eeh_pe_dev_traverse(pe, eeh_report_resume, NULL);
534 /* Clear recovery mode */
535 eeh_pe_state_clear(pe, EEH_PE_RECOVERING);
541 * eeh_reset_device - Perform actual reset of a pci slot
543 * @bus: PCI bus corresponding to the isolcated slot
545 * This routine must be called to do reset on the indicated PE.
546 * During the reset, udev might be invoked because those affected
547 * PCI devices will be removed and then added.
549 static int eeh_reset_device(struct eeh_pe *pe, struct pci_bus *bus)
551 struct pci_bus *frozen_bus = eeh_pe_bus_get(pe);
552 struct timeval tstamp;
553 int cnt, rc, removed = 0;
555 /* pcibios will clear the counter; save the value */
556 cnt = pe->freeze_count;
560 * We don't remove the corresponding PE instances because
561 * we need the information afterwords. The attached EEH
562 * devices are expected to be attached soon when calling
563 * into pcibios_add_pci_devices().
565 eeh_pe_state_mark(pe, EEH_PE_KEEP);
567 pci_lock_rescan_remove();
568 pcibios_remove_pci_devices(bus);
569 pci_unlock_rescan_remove();
570 } else if (frozen_bus) {
571 eeh_pe_dev_traverse(pe, eeh_rmv_device, &removed);
575 * Reset the pci controller. (Asserts RST#; resets config space).
576 * Reconfigure bridges and devices. Don't try to bring the system
577 * up if the reset failed for some reason.
579 * During the reset, it's very dangerous to have uncontrolled PCI
580 * config accesses. So we prefer to block them. However, controlled
581 * PCI config accesses initiated from EEH itself are allowed.
583 rc = eeh_reset_pe(pe);
587 pci_lock_rescan_remove();
590 eeh_ops->configure_bridge(pe);
591 eeh_pe_restore_bars(pe);
594 * If it's PHB PE, the frozen state on all available PEs should have
595 * been cleared by the PHB reset. Otherwise, we unfreeze the PE and its
596 * child PEs because they might be in frozen state.
598 if (!(pe->type & EEH_PE_PHB)) {
599 rc = eeh_clear_pe_frozen_state(pe, false);
604 /* Give the system 5 seconds to finish running the user-space
605 * hotplug shutdown scripts, e.g. ifdown for ethernet. Yes,
606 * this is a hack, but if we don't do this, and try to bring
607 * the device up before the scripts have taken it down,
608 * potentially weird things happen.
611 pr_info("EEH: Sleep 5s ahead of complete hotplug\n");
615 * The EEH device is still connected with its parent
616 * PE. We should disconnect it so the binding can be
617 * rebuilt when adding PCI devices.
619 eeh_pe_traverse(pe, eeh_pe_detach_dev, NULL);
620 pcibios_add_pci_devices(bus);
621 } else if (frozen_bus && removed) {
622 pr_info("EEH: Sleep 5s ahead of partial hotplug\n");
625 eeh_pe_traverse(pe, eeh_pe_detach_dev, NULL);
626 pcibios_add_pci_devices(frozen_bus);
628 eeh_pe_state_clear(pe, EEH_PE_KEEP);
631 pe->freeze_count = cnt;
633 pci_unlock_rescan_remove();
637 /* The longest amount of time to wait for a pci device
638 * to come back on line, in seconds.
640 #define MAX_WAIT_FOR_RECOVERY 300
642 static void eeh_handle_normal_event(struct eeh_pe *pe)
644 struct pci_bus *frozen_bus;
646 enum pci_ers_result result = PCI_ERS_RESULT_NONE;
648 frozen_bus = eeh_pe_bus_get(pe);
650 pr_err("%s: Cannot find PCI bus for PHB#%d-PE#%x\n",
651 __func__, pe->phb->global_number, pe->addr);
655 eeh_pe_update_time_stamp(pe);
657 if (pe->freeze_count > eeh_max_freezes)
658 goto excess_failures;
659 pr_warn("EEH: This PCI device has failed %d times in the last hour\n",
662 /* Walk the various device drivers attached to this slot through
663 * a reset sequence, giving each an opportunity to do what it needs
664 * to accomplish the reset. Each child gets a report of the
665 * status ... if any child can't handle the reset, then the entire
666 * slot is dlpar removed and added.
668 * When the PHB is fenced, we have to issue a reset to recover from
669 * the error. Override the result if necessary to have partially
670 * hotplug for this case.
672 pr_info("EEH: Notify device drivers to shutdown\n");
673 eeh_pe_dev_traverse(pe, eeh_report_error, &result);
674 if ((pe->type & EEH_PE_PHB) &&
675 result != PCI_ERS_RESULT_NONE &&
676 result != PCI_ERS_RESULT_NEED_RESET)
677 result = PCI_ERS_RESULT_NEED_RESET;
679 /* Get the current PCI slot state. This can take a long time,
680 * sometimes over 300 seconds for certain systems.
682 rc = eeh_ops->wait_state(pe, MAX_WAIT_FOR_RECOVERY*1000);
683 if (rc < 0 || rc == EEH_STATE_NOT_SUPPORT) {
684 pr_warn("EEH: Permanent failure\n");
688 /* Since rtas may enable MMIO when posting the error log,
689 * don't post the error log until after all dev drivers
690 * have been informed.
692 pr_info("EEH: Collect temporary log\n");
693 eeh_slot_error_detail(pe, EEH_LOG_TEMP);
695 /* If all device drivers were EEH-unaware, then shut
696 * down all of the device drivers, and hope they
697 * go down willingly, without panicing the system.
699 if (result == PCI_ERS_RESULT_NONE) {
700 pr_info("EEH: Reset with hotplug activity\n");
701 rc = eeh_reset_device(pe, frozen_bus);
703 pr_warn("%s: Unable to reset, err=%d\n",
709 /* If all devices reported they can proceed, then re-enable MMIO */
710 if (result == PCI_ERS_RESULT_CAN_RECOVER) {
711 pr_info("EEH: Enable I/O for affected devices\n");
712 rc = eeh_pci_enable(pe, EEH_OPT_THAW_MMIO);
717 result = PCI_ERS_RESULT_NEED_RESET;
719 pr_info("EEH: Notify device drivers to resume I/O\n");
720 eeh_pe_dev_traverse(pe, eeh_report_mmio_enabled, &result);
724 /* If all devices reported they can proceed, then re-enable DMA */
725 if (result == PCI_ERS_RESULT_CAN_RECOVER) {
726 pr_info("EEH: Enabled DMA for affected devices\n");
727 rc = eeh_pci_enable(pe, EEH_OPT_THAW_DMA);
732 result = PCI_ERS_RESULT_NEED_RESET;
735 * We didn't do PE reset for the case. The PE
736 * is still in frozen state. Clear it before
739 eeh_pe_state_clear(pe, EEH_PE_ISOLATED);
740 result = PCI_ERS_RESULT_RECOVERED;
744 /* If any device has a hard failure, then shut off everything. */
745 if (result == PCI_ERS_RESULT_DISCONNECT) {
746 pr_warn("EEH: Device driver gave up\n");
750 /* If any device called out for a reset, then reset the slot */
751 if (result == PCI_ERS_RESULT_NEED_RESET) {
752 pr_info("EEH: Reset without hotplug activity\n");
753 rc = eeh_reset_device(pe, NULL);
755 pr_warn("%s: Cannot reset, err=%d\n",
760 pr_info("EEH: Notify device drivers "
761 "the completion of reset\n");
762 result = PCI_ERS_RESULT_NONE;
763 eeh_pe_dev_traverse(pe, eeh_report_reset, &result);
766 /* All devices should claim they have recovered by now. */
767 if ((result != PCI_ERS_RESULT_RECOVERED) &&
768 (result != PCI_ERS_RESULT_NONE)) {
769 pr_warn("EEH: Not recovered\n");
773 /* Tell all device drivers that they can resume operations */
774 pr_info("EEH: Notify device driver to resume\n");
775 eeh_pe_dev_traverse(pe, eeh_report_resume, NULL);
781 * About 90% of all real-life EEH failures in the field
782 * are due to poorly seated PCI cards. Only 10% or so are
783 * due to actual, failed cards.
785 pr_err("EEH: PHB#%d-PE#%x has failed %d times in the\n"
786 "last hour and has been permanently disabled.\n"
787 "Please try reseating or replacing it.\n",
788 pe->phb->global_number, pe->addr,
793 pr_err("EEH: Unable to recover from failure from PHB#%d-PE#%x.\n"
794 "Please try reseating or replacing it\n",
795 pe->phb->global_number, pe->addr);
798 eeh_slot_error_detail(pe, EEH_LOG_PERM);
800 /* Notify all devices that they're about to go down. */
801 eeh_pe_dev_traverse(pe, eeh_report_failure, NULL);
803 /* Mark the PE to be removed permanently */
804 eeh_pe_state_mark(pe, EEH_PE_REMOVED);
807 * Shut down the device drivers for good. We mark
808 * all removed devices correctly to avoid access
809 * the their PCI config any more.
812 eeh_pe_dev_mode_mark(pe, EEH_DEV_REMOVED);
814 pci_lock_rescan_remove();
815 pcibios_remove_pci_devices(frozen_bus);
816 pci_unlock_rescan_remove();
820 static void eeh_handle_special_event(void)
822 struct eeh_pe *pe, *phb_pe;
824 struct pci_controller *hose;
830 rc = eeh_ops->next_error(&pe);
833 case EEH_NEXT_ERR_DEAD_IOC:
834 /* Mark all PHBs in dead state */
835 eeh_serialize_lock(&flags);
837 /* Purge all events */
838 eeh_remove_event(NULL, true);
840 list_for_each_entry(hose, &hose_list, list_node) {
841 phb_pe = eeh_phb_pe_get(hose);
842 if (!phb_pe) continue;
844 eeh_pe_state_mark(phb_pe, EEH_PE_ISOLATED);
847 eeh_serialize_unlock(flags);
850 case EEH_NEXT_ERR_FROZEN_PE:
851 case EEH_NEXT_ERR_FENCED_PHB:
852 case EEH_NEXT_ERR_DEAD_PHB:
853 /* Mark the PE in fenced state */
854 eeh_serialize_lock(&flags);
856 /* Purge all events of the PHB */
857 eeh_remove_event(pe, true);
859 if (rc == EEH_NEXT_ERR_DEAD_PHB)
860 eeh_pe_state_mark(pe, EEH_PE_ISOLATED);
862 eeh_pe_state_mark(pe,
863 EEH_PE_ISOLATED | EEH_PE_RECOVERING);
865 eeh_serialize_unlock(flags);
868 case EEH_NEXT_ERR_NONE:
871 pr_warn("%s: Invalid value %d from next_error()\n",
877 * For fenced PHB and frozen PE, it's handled as normal
878 * event. We have to remove the affected PHBs for dead
881 if (rc == EEH_NEXT_ERR_FROZEN_PE ||
882 rc == EEH_NEXT_ERR_FENCED_PHB) {
883 eeh_handle_normal_event(pe);
884 eeh_pe_state_clear(pe, EEH_PE_RECOVERING);
886 pci_lock_rescan_remove();
887 list_for_each_entry(hose, &hose_list, list_node) {
888 phb_pe = eeh_phb_pe_get(hose);
890 !(phb_pe->state & EEH_PE_ISOLATED) ||
891 (phb_pe->state & EEH_PE_RECOVERING))
894 /* Notify all devices to be down */
895 bus = eeh_pe_bus_get(phb_pe);
896 eeh_pe_dev_traverse(pe,
897 eeh_report_failure, NULL);
898 pcibios_remove_pci_devices(bus);
900 pci_unlock_rescan_remove();
904 * If we have detected dead IOC, we needn't proceed
905 * any more since all PHBs would have been removed
907 if (rc == EEH_NEXT_ERR_DEAD_IOC)
909 } while (rc != EEH_NEXT_ERR_NONE);
913 * eeh_handle_event - Reset a PCI device after hard lockup.
916 * While PHB detects address or data parity errors on particular PCI
917 * slot, the associated PE will be frozen. Besides, DMA's occurring
918 * to wild addresses (which usually happen due to bugs in device
919 * drivers or in PCI adapter firmware) can cause EEH error. #SERR,
920 * #PERR or other misc PCI-related errors also can trigger EEH errors.
922 * Recovery process consists of unplugging the device driver (which
923 * generated hotplug events to userspace), then issuing a PCI #RST to
924 * the device, then reconfiguring the PCI config space for all bridges
925 * & devices under this slot, and then finally restarting the device
926 * drivers (which cause a second set of hotplug events to go out to
929 void eeh_handle_event(struct eeh_pe *pe)
932 eeh_handle_normal_event(pe);
934 eeh_handle_special_event();